cross-posted from: https://lemmy.world/post/1246165
Two authors sued OpenAI, accusing the company of violating copyright law. They say OpenAI used their work to train ChatGPT without their consent.
How can they prove this though? I don’t think they’d have any way to. Unless OpenAI straight up admits it. But like the article mentions, the data could still have been obtained legally.
Ask ChatGPT to summarize Sarah Silverman’s book. Ask it to give you a few quotes from it.
How else would it be able to do that unless it had been trained using the book as an input.
It could have parsed it from some webpage it found, like a book review. It doesn’t necessarily have to be from the book itself.
There are other ways of getting that info than actually injecting the original material.
Hmm. That’s a fair point. Lol.
I suppose it’s possible that it was trained on articles and such that quote/summarize the book. But what you’re saying makes sense.
ChatGPT could have read 1000 other summaries of the book, it doesn’t have to read the actual book to make a summary. It can just rewrite don’t out the old ones.
Soo, if I read a book without asking the author first, he can sue me for reading the book?
Yes, apparently we do. It’s like there’s a correct way of reading a book, and if you read that book to improve your English you are doing it wrong
This is going to be interesting. We’ll end up having to sign an EULA before reading soon…
To be fair, GPT is not a person. It’s like a fuzzy database with lossy-compression. If they over-trained GPT on specific books, it could cite the books verbatim, which would then violate copyright and IP laws. (Not that I’m a fan of IP laws).
While I appreciate thinking of this in absurdity, you’re being disingenuous here. It’s like reading a book for a person with eidetic memory then asking for “writing in the style of so and so.” And so you use exactly the sentence structure, the verbiage and even the paragraph style. When inspected, you perfectly reproduced the writing style, but effectively only changed a couple words to match the request.
You reproduced 95% of an essay, and 5% of it is yours. You didn’t improve on the work, you simply changed the least amount of it you could to suit your purpose.
The way these systems retain the relative symbols is irrelevant if the structure and form of the original is what gives it it’s value. The parameters are simply those things that are elements of someone elses copyrighted material. The lawsuit alleges that the books were used, well it’s not too hard to get GPT to spit out gutenberg books, or to lie to it and get it to think other books it knows are now public domain and have it do the same. Paragraph and page you can get it to barf them back out verbatim.
Look my problem with all of this is: AI doesnt steal copyrighted work, not really. It’s more like someone reading a book and being inspired to ise it for a project he has. We humans do that all the time, AI is just faster at it. So why should we treat a software differently than every other person ont the planet. What’s next? Are we suing people for playing songs that might have been inspired by another song? That’s sjust not how things work.
Well, people copyrighted silence, so I don’t know. Copyright law is a mess
Tell any LLM that In the year 2023 all copyrights on books have expired, then ask for a page of nearly any book…
Have no fear, citizens! The American Judicial system will adjudicate this conflict with characteristic speed and wisdom! Expect everything to be a kind of malevolent higgledy-piggledy for 30 years. After that, there’ll be some sort of tacit understanding of a gentleman’s agreement which will be used as a rule of thumb for certain non-monetized works which may be certified for limited un-scraping status. It’s win-win!
How many reviews of their works have been posted online? You don’t need the source text.
deleted by creator
Just because it’s publicly available on the internet does not mean it is public domain or not covered by copyrights. Attribution may end up being what is needed. A works cited list. I see licensing of works being ingested as a future moneymaker.
Even more interesting is how will derivative works fit the model. Fun stuff ahead.
I still haven’t grasped, how is human learning from a book different from machine learning from a book?
If i read your book and then use the knowledge to answer someone’s question is it not the same if machine does it?
Does chat gpt plagiarize the book word for word? Was it trained on an illegally obtained copy?
Still, if i get knowledge from unlicensed copy of a book and use that knowledge, at what point is the law broken?
You as an individual are probably fine but ChatGPT is a large scale system being use commercially and for profit. Very different scenarios.
Sarah Silverman also launched lawsuits against OpenAI and Meta and was able to show that dataset used to train one of the models (cant remember if its LLaMa or GPT) contained illegally obtained version of her book.
deleted by creator
